Your Name Here: Sheraz Choudhary¶

Your TF Name: Joe Palin¶

Graduate or Undergraduate Credit: Graduate¶

Image Exif Data Analyzer¶

import pandas as pd
import numpy as np
import os, sys
import PIL.Image
from PIL import Image
import PIL.ExifTags
from PIL.ExifTags import TAGS, GPSTAGS

Set Arguments for the images to be scanned:¶

#directory to scan 
directory="C:/Users/shera/Pictures/2018"
incl_ext=('JPG', 'jpg', 'JPEG', 'jpeg', 'TIFF', 'tiff', 'PNG', 'png')  # files to include in file scan

Define Functions to Scan Directory and Extract EXIF:¶

def exif_to_df(path):
    """
    Given the full path to an image this function extracts the Exif labels and values of the image.  
    A pandas dataframe is returned with the labels as column names.
    """
    img = Image.open(path)  #open picture at input path 
    
    try:
        imgExif = {TAGS[label]: value  #list comprehension to create dictionary of Exif label in PIL.ExifTags
                   for label, value in img._getexif().items()
                   if label in TAGS}
    except:  #if there is no Exif data return none
        return None
    
    imgExif.pop('MakerNote', None)  #remove MakerNote because it differ with each camera and data format is proprietary
    imgExif.pop('UserComment', None)  #this field in the exif is not used and contains dummy data

    imgExif['ImagePath'] = path
    
    imgExifDf = pd.DataFrame([imgExif])  #convert dictionary to dataframe (https://stackoverflow.com/questions/18837262/convert-python-dict-into-a-dataframe)
    
    return imgExifDf

def scan_exif(dir):
    """
    Scans all files in the input directory (valid path to folder in string format).
    The path of the file is used to call the exif_to_df function to get a dataframe with the Exif values.
    """
    all_df = pd.DataFrame()  #this df will hold all image df (complication is that each image df can have some diff and some same columns)

    for path, dirs, files in os.walk(dir, topdown=True):  #https://gist.github.com/Grimthorr/8ea07f43cebeb4156e54   #https://www.tutorialspoint.com/python/os_walk.htm   #https://stackoverflow.com/questions/30444105/better-way-to-find-absolute-paths-during-os-walk
        for file in files:
            if file.endswith(incl_ext):  #only include file types with exif data (specified as tuple below import statements)
                pathfile=os.path.join(os.path.abspath(path), file)  #absolute path and filename concatenated  #https://stackoverflow.com/questions/30444105/better-way-to-find-absolute-paths-during-os-walk   #https://docs.python.org/2/library/stat.html and https://docs.python.org/3/library/os.html
                img_df = exif_to_df(pathfile)
                all_df = all_df.append(img_df, ignore_index=True)  #append image df to all_df (https://stackoverflow.com/questions/43720337/pandatables-and-exif-adding-columns-as-needed)

    return all_df

Call Functions to create Exif Dataframe and CSV:¶

# Read all images in directory, create exif dataframe, write to csv file
# Takes around 7-10 min to scan a directory with 13000 images (create backup_df incase corrupt exif_df during cleaning)
exif_df = scan_exif(directory)
exif_df.to_csv('exif_file.csv')

# # code used during testing:
# backup_df = scan_exif(directory)
# exif_df = backup_df
# exif_df.to_csv('exif_file.csv')
# exif_df = backup_df.copy()

pd.set_option('display.max_columns', 150)
pd.set_option('display.max_rows', 1000)
display (exif_df.shape)
exif_df.head()

(13213, 75)

Let the Data Cleanup Begin¶

Research columns and then remove all that deemed unnecessary¶

exif_df = exif_df.drop(['ApertureValue', 'Artist', 'BitsPerSample', 'BodySerialNumber', 'CameraOwnerName', 'ColorSpace', 
                        'ComponentsConfiguration', 'CompressedBitsPerPixel', 'Contrast', 'Copyright', 'CustomRendered', 
                        'DigitalZoomRatio', 'ExifImageHeight', 'ExifImageWidth', 'ExifInteroperabilityOffset', 'ExifOffset', 
                        'ExifVersion', 'FileSource', 'FlashPixVersion', 'FocalPlaneResolutionUnit', 'FocalPlaneXResolution', 
                        'FocalPlaneYResolution', 'GPSInfo', 'ImageDescription', 'ImageLength', 'ImageUniqueID', 'ImageWidth', 
                        'LensMake', 'LensModel', 'LensSerialNumber', 'LightSource', 'LensSpecification', 'MaxApertureValue', 
                        'Orientation', 'PhotometricInterpretation', 'PrintImageMatching', 'ProcessingSoftware', 
                        'ResolutionUnit', 'SamplesPerPixel', 'Saturation', 'SceneCaptureType', 'SceneType', 'SensingMethod', 
                        'Sharpness', 'ShutterSpeedValue', 'SubjectLocation', 'SubsecTime', 'SubsecTimeDigitized', 
                        'SubsecTimeOriginal', 'XResolution', 'YCbCrPositioning', 'YResolution'], axis=1)

Fill some missing values, simplify values and delete rows not needed¶

# Remove row where 'Model' is null or iPhone
print("Before:", exif_df.shape)
exif_df = exif_df.dropna(subset=['Model'])
exif_df = exif_df[exif_df['Make'] != 'Apple']
print("After:", exif_df.shape)

Before: (13213, 23)
After: (13194, 23)

# create mask for canon and replace nulls in 'FocalLengthIn35mmFilm' with appropriate values
canon_mask = exif_df['Model'].str.contains('Canon')

exif_df['FocalLengthIn35mmFilm'][canon_mask] = exif_df['FocalLength']  # canon is a 35mm camera so these are equivalent

# clean-up df rows from externally modified sources using the 'Software' column
print("Before:", exif_df.shape)
exif_df['Software'] = exif_df['Software'].replace(np.nan, 'Unknown')  # Canon does not populate 'Software'
exif_df = exif_df[exif_df['Software'].str.contains('Picasa') == False]
exif_df = exif_df[exif_df['Software'].str.contains('Adobe ') == False]
exif_df = exif_df[exif_df['Software'].str.contains('Windows ') == False]
exif_df = exif_df[exif_df['Software'].str.contains('marlin') == False]
print("After:", exif_df.shape)

Before: (13194, 23)
After: (12630, 23)

# Simplify flash values to 0 or 1 based on if the flash fired or not
flash_y_mask = exif_df['Flash'].isin([9, 13, 15])
flash_n_mask = exif_df['Flash'] == 16

exif_df['Flash'][flash_y_mask] = 1  # 9,13,15 all mean flash fired
exif_df['Flash'][flash_n_mask] = 0  # 16 means flash did not fire

C:\Users\shera\Anaconda3\lib\site-packages\ipykernel_launcher.py:6: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy

# create masks for filenames (filenames roughly correlate with picture content--estimate 90% correctly labeled)
baby_some_mask = exif_df['ImagePath'].str.contains('Zak')
baby_all_mask = exif_df['ImagePath'].str.contains('ZAK')
copy_mask = exif_df['ImagePath'].str.contains('Copy')
random_mask = exif_df['ImagePath'].str.contains('Random')

exif_df['Subject'] = np.nan

exif_df['Subject'][baby_some_mask] = 'SomeBaby'
exif_df['Subject'][baby_all_mask] = 'AllBaby'
exif_df['Subject'][copy_mask] = 'Copy'
exif_df['Subject'][random_mask] = 'Random'
exif_df['Subject'] = exif_df['Subject'].replace(np.nan, 'Misc')

C:\Users\shera\Anaconda3\lib\site-packages\ipykernel_launcher.py:9: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
  if __name__ == '__main__':
C:\Users\shera\Anaconda3\lib\site-packages\ipykernel_launcher.py:10: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
  # Remove the CWD from sys.path while we load stuff.
C:\Users\shera\Anaconda3\lib\site-packages\ipykernel_launcher.py:11: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
  # This is added back by InteractiveShellApp.init_path()
C:\Users\shera\Anaconda3\lib\site-packages\ipykernel_launcher.py:12: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
  if sys.path[0] == '':

Convert tuples to decimal and calculate APEX values¶

def val_to_num (val):
    if isinstance(val, tuple):  #convert only tuples  (eg. (280, 100) will be converted to 2.8)
        decimal = val[0]/val[1]
        return round(decimal, 5)
    else:
        return val

# convert all tuple entries to decimal floats
exif_df = exif_df.applymap(lambda val: val_to_num(val))

import math

def apex_vals(fnum, shutter, iso):
    av = round(math.log(fnum**2, 2), 3)
    tv = round(math.log(1/shutter, 2), 3)
    sv = round(math.log((iso/3.125), 2), 3)
    ev = round(av + tv, 3)
    bv = round(ev - sv, 3)
    return (av, tv, sv, bv)


def exposure_val(fnum, shutter, iso):
    ev = math.log((fnum**2 / shutter), 2)
    ev_iso = ev - math.log((iso/100), 2)
    ev_iso = round(ev_iso, 3)
    return ev_iso


def dof_val(fnum, focal, focal35):
    
    c = 0.03 * (focal / focal35)  # calculate circle of confusion based on 0.03 for 35mm sensor
        
    # estimate subject distance based on focallength35
    # custom formula to make nexus distance 1 meter and Canon at 200 distance around 3 meters
    s = focal35 * 2.7196 * (1/c**.5)
    
    h = (focal**2/(fnum*c)) + focal
    dn = (s*(h - focal)) / (h + s - 2*focal)
    df = (s*(h - focal)) / (h - s)
    dof = (df - dn) / 1000  # all units were in mm so convert to meters 
    return dof
   
    
# Test functions to calculate av, tv, sv, bv, ev
print("Ev:", exposure_val(2.8, 0.008, 1600))
print("Av, Tv, Sv, Bv:", apex_vals(2.8, 0.008, 1600))
print("dof:", dof_val(2, 4.6, 26))

Ev: 5.937
Av, Tv, Sv, Bv: (2.971, 6.966, 9.0, 0.937)
dof: 1.2293720572071598

# convert exposure values to APEX system 
# eg. f2.8 lets in 2x the light of 4.0 and 4x the light of 5.6 (in APEX 2.8, 4.0, 5.6 correspond to 3, 4, 5)
exif_df['Av'] = exif_df.apply(lambda row: apex_vals(row['FNumber'], row['ExposureTime'], row['ISOSpeedRatings'])[0], axis=1)
exif_df['Tv'] = exif_df.apply(lambda row: apex_vals(row['FNumber'], row['ExposureTime'], row['ISOSpeedRatings'])[1], axis=1)
exif_df['Sv'] = exif_df.apply(lambda row: apex_vals(row['FNumber'], row['ExposureTime'], row['ISOSpeedRatings'])[2], axis=1)
exif_df['Bv'] = exif_df.apply(lambda row: apex_vals(row['FNumber'], row['ExposureTime'], row['ISOSpeedRatings'])[3], axis=1)

# add calculated dof to the dataframe
exif_df['Dof'] = exif_df.apply(lambda row: dof_val(row['FNumber'], row['FocalLength'], row['FocalLengthIn35mmFilm']), axis=1)

Bin the data and create new feature columns¶

exif_df['DateTime'] = pd.to_datetime(exif_df['DateTime'], format='%Y:%m:%d %H:%M:%S') #2018:03:01 07:30:17

# Create bins based on hour of day (code modified from Lecture 11 code)
# exif_df['Hour'] = exif_df['DateTime'].apply(lambda x: round(x.hour+(x.minute/60),1))  #if want more precise hour
exif_df['Hour'] = exif_df['DateTime'].apply(lambda x: x.hour)  #if want more precise hour

hr_bins = [-1, 7, 9, 17, 19, 23]
bin_names = ['LateNight', 'Morning', 'Daytime', 'Evening', 'Night']  # text-based bins
# bin_names = [4, 0, 1, 2, 3]  # numerical bins

exif_df['TimeBin'] = pd.cut(exif_df['Hour'], bins=hr_bins, labels=bin_names)
exif_df['TimeBin'] = exif_df['TimeBin'].astype('str')  # convert from category to str so LabelEncoder can be used
# exif_df['TimeBin'] = pd.to_numeric(exif_df['TimeBin'])  # comment out if using text-based bins

# Create bins based on hour of day (code modified from Lecture 11 code)
bv_bins = [-15, -5, 0, 5, 9, 15]
bv_bin_names = ['Dim Indoor', 'Indoor', 'Dusk', 'Weak Sun', 'Strong Sun']  # text-based bins
# bin_names = [0, 1, 2, 3, 4]  # numerical bins

exif_df['BrightnessBin'] = pd.cut(exif_df['Bv'], bins=bv_bins, labels=bv_bin_names)
exif_df['BrightnessBin'] = exif_df['BrightnessBin'].astype('str')  # convert from category to str so LabelEncoder can be used

flash_y_mask = exif_df['Flash'] == 1
exif_df['BrightnessBin'][flash_y_mask] = 'Flash'  # override brightness bin if flash was used
# exif_df['TimeBin'] = pd.to_numeric(exif_df['TimeBin'])  # comment out if using text-based bins

C:\Users\shera\Anaconda3\lib\site-packages\ipykernel_launcher.py:10: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
  # Remove the CWD from sys.path while we load stuff.

Label Encoding for Text columns¶

# code adapted from Lecture 9 code
from sklearn.preprocessing import OneHotEncoder, LabelEncoder

# instantiate the encoder
labels = LabelEncoder()

# fit label encoder to text columns
labels_subject = labels.fit_transform(exif_df['Subject'])
labels_model = labels.fit_transform(exif_df['Model'])
labels_timebin = labels.fit_transform(exif_df['TimeBin'])
labels_bvbin = labels.fit_transform(exif_df['BrightnessBin'])

# add the columns with out of labelencoder fit 
exif_df['Subject_L'] = labels_subject
exif_df['Model_L'] = labels_model
exif_df['TimeBin_L'] = labels_timebin
exif_df['BrightnessBin_L'] = labels_bvbin

Remove additional columns no longer need and reorder the columns¶

exif_df = exif_df.drop(['BrightnessValue', 'DateTimeDigitized', 'DateTimeOriginal', 'ExposureMode', 'FocalLength', 'Make', 
                        'MeteringMode', 'Software', 'SubjectDistance', 'SubjectDistanceRange', 'WhiteBalance'], axis=1)

exif_df = exif_df[['Subject', 'Subject_L', 'ImagePath', 'DateTime', 'Hour', 'TimeBin', 'TimeBin_L', 'Model', 'Model_L', 
                   'FocalLengthIn35mmFilm', 'ExposureProgram', 'ExposureBiasValue', 'Flash', 'Dof','FNumber', 'Av', 
                   'ExposureTime', 'Tv', 'ISOSpeedRatings', 'Sv', 'Bv', 'BrightnessBin', 'BrightnessBin_L']]

exif_verbose_df = exif_df.copy()

exif_df = exif_df.drop(['FNumber', 'ExposureTime', 'ISOSpeedRatings'], axis=1)

Finally, remove any remaining nulls and test shape of df¶

From starting shape of (13166, 75) to cleaned shape of (12571, 11)¶

exif_df = exif_df.dropna()
exif_verbose_df = exif_verbose_df.dropna()

exif_df.isna().any()

Subject                  False
Subject_L                False
ImagePath                False
DateTime                 False
Hour                     False
TimeBin                  False
TimeBin_L                False
Model                    False
Model_L                  False
FocalLengthIn35mmFilm    False
ExposureProgram          False
ExposureBiasValue        False
Flash                    False
Dof                      False
Av                       False
Tv                       False
Sv                       False
Bv                       False
BrightnessBin            False
BrightnessBin_L          False
dtype: bool

exif_df.dtypes

Subject                          object
Subject_L                         int64
ImagePath                        object
DateTime                 datetime64[ns]
Hour                            float64
TimeBin                          object
TimeBin_L                         int64
Model                            object
Model_L                           int64
FocalLengthIn35mmFilm           float64
ExposureProgram                 float64
ExposureBiasValue               float64
Flash                           float64
Dof                             float64
Av                              float64
Tv                              float64
Sv                              float64
Bv                              float64
BrightnessBin                    object
BrightnessBin_L                   int64
dtype: object

exif_df = exif_df.reset_index(drop=True)

print("Shape of Cleaned Df:", exif_df.shape)
display(exif_df.tail())

Shape of Cleaned Df: (12616, 20)

exif_verbose_df.to_csv('exif_cleaned.csv')

Let the Analysis Begin¶

(no kidding about 70% of time being spent on cleanup)¶

exif_df.corr()

exif_df.describe()

Visualize the data¶

import matplotlib.pyplot as plt
import seaborn as sns

graph = sns.pairplot(exif_df)
graph.savefig('exif_sns.png')
plt.show()

import plotly
plotly.offline.init_notebook_mode(connected=True)
import plotly.graph_objs as go

import cufflinks as cf
cf.go_offline() #from Lecture 4 (otherwise causes user authentication error)

#load the data
byHour = exif_verbose_df['ImagePath'].groupby(exif_verbose_df['Hour']).count()

#Got syntax for cufflinks from: https://plot.ly/ipython-notebooks/cufflinks/
byHour.iplot(kind='bar', 
             xTitle='Hour of Day', 
             yTitle="Number of Photos", 
             title='Number of Photos by Hour of Day', 
             filename='hour_graph')

#load the data
byModel = exif_verbose_df['ImagePath'].groupby(exif_verbose_df['Model']).count()

#Got syntax for cufflinks from: https://plot.ly/ipython-notebooks/cufflinks/
byModel.iplot(kind='bar', 
             xTitle='Camera Model', 
             yTitle="Number of Photos", 
             title='Number of Photos by Camera Model', 
             filename='camera_graph')

#load the data
bySubject = exif_verbose_df['ImagePath'].groupby(exif_verbose_df['Subject']).count()

#Got syntax for cufflinks from: https://plot.ly/ipython-notebooks/cufflinks/
bySubject.iplot(kind='bar', 
             xTitle='Subject', 
             yTitle="Number of Photos", 
             title='Number of Photos by Subject', 
             filename='subject_graph')

#load the data
byFocal = exif_verbose_df['ImagePath'].groupby(exif_verbose_df['FocalLengthIn35mmFilm']).count()

#Got syntax for cufflinks from: https://plot.ly/ipython-notebooks/cufflinks/
byFocal.iplot(kind='bar', 
             xTitle='35mm Focal Length', 
             yTitle="Number of Photos", 
             title='Number of Photos by Focal Length', 
             filename='focal_graph')

#load the data
byAperture = exif_verbose_df['ImagePath'].groupby(exif_verbose_df['FNumber']).count()

#Got syntax for cufflinks from: https://plot.ly/ipython-notebooks/cufflinks/
byAperture.iplot(kind='bar', 
             xTitle='Aperture', 
             yTitle="Number of Photos", 
             title='Number of Photos by Aperture', 
             filename='aperture_graph')

#load the data
byShutter = exif_verbose_df['ImagePath'].groupby(exif_verbose_df['ExposureTime']).count()

#Got syntax for cufflinks from: https://plot.ly/ipython-notebooks/cufflinks/
byShutter.iplot(kind='bar', 
             xTitle='Shutter', 
             yTitle="Number of Photos", 
             title='Number of Photos by Shutter', 
             filename='shutter_graph')

#load the data
byISO = exif_verbose_df['ImagePath'].groupby(exif_verbose_df['ISOSpeedRatings']).count()

#Got syntax for cufflinks from: https://plot.ly/ipython-notebooks/cufflinks/
byISO.iplot(kind='bar', 
             xTitle='ISO Speed', 
             yTitle="Number of Photos", 
             title='Number of Photos by ISO Speed', 
             filename='iso_graph')

#load the data
byBrightness = exif_verbose_df['ImagePath'].groupby(exif_verbose_df['BrightnessBin']).count()

#Got syntax for cufflinks from: https://plot.ly/ipython-notebooks/cufflinks/
byBrightness.iplot(kind='bar', 
             xTitle='Brightness', 
             yTitle="Number of Photos", 
             title='Number of Photos by Brightness', 
             filename='brightness_graph')

#load the data
byExposureBias = exif_verbose_df['ImagePath'].groupby(exif_verbose_df['ExposureBiasValue']).count()

#Got syntax for cufflinks from: https://plot.ly/ipython-notebooks/cufflinks/
byExposureBias.iplot(kind='bar', 
             xTitle='Exposure Bias', 
             yTitle="Number of Photos", 
             title='Number of Photos by Exposure Bias', 
             filename='exposurebias_graph')

Specify features and targets then split into training and testing dfs¶

from sklearn.model_selection import train_test_split

X = exif_df[['Hour', 'TimeBin_L', 'Model_L', 'FocalLengthIn35mmFilm', 'ExposureProgram', 'ExposureBiasValue', 'Flash', 
            'Dof', 'Av', 'Tv', 'Sv', 'Bv', 'BrightnessBin_L']]
y = exif_df['Subject']

# all = exif_df[['Subject_L', 'Hour', 'TimeBin_L', 'Model_L', 'FocalLengthIn35mmFilm', 'ExposureProgram', 'ExposureBiasValue', 
#                'Flash', 'Dof', 'Av', 'Tv', 'Sv', 'Bv', 'BrightnessBin_L']]

display(X.head(3))
display(y.head(3))

X_train,X_test,y_train,y_test = train_test_split(X, y, test_size=0.25, random_state=81, stratify=y)

0    AllBaby
1    AllBaby
2    AllBaby
Name: Subject, dtype: object

Let's try some models¶

Decision Tree¶

from IPython.display import display, Image
from sklearn.tree import DecisionTreeClassifier
from sklearn import tree
from sklearn.metrics import confusion_matrix
from sklearn.metrics import classification_report
from sklearn.metrics import accuracy_score
import pydotplus

tree_clf = DecisionTreeClassifier(max_depth=5, min_samples_leaf=250, random_state=81)
tree_clf.fit(X_train, y_train)

DecisionTreeClassifier(class_weight=None, criterion='gini', max_depth=5,
            max_features=None, max_leaf_nodes=None,
            min_impurity_decrease=0.0, min_impurity_split=None,
            min_samples_leaf=250, min_samples_split=2,
            min_weight_fraction_leaf=0.0, presort=False, random_state=81,
            splitter='best')

# Decided to use a decision tree
# 0.8 is the best score with an overfitted tree but pruning it to a more manageable level
tree_clf.score(X_test,y_test)

0.6712111604311984

tree_confusion_matrix = confusion_matrix(y_test, tree_clf.predict(X_test))
print(tree_confusion_matrix)
#sns.heatmap(tree_confusion_matrix)

[[1870    0   11    0   12]
 [  19    0    6    0    0]
 [ 131    0  119    0   10]
 [  79    0   11    0   32]
 [ 684    0   42    0  128]]

print(classification_report(y_test, tree_clf.predict(X_test)))  # does not display properly with display()

             precision    recall  f1-score   support

    AllBaby       0.67      0.99      0.80      1893
       Copy       0.00      0.00      0.00        25
       Misc       0.63      0.46      0.53       260
     Random       0.00      0.00      0.00       122
   SomeBaby       0.70      0.15      0.25       854

avg / total       0.65      0.67      0.59      3154

C:\Users\shera\Anaconda3\lib\site-packages\sklearn\metrics\classification.py:1135: UndefinedMetricWarning:

Precision and F-score are ill-defined and being set to 0.0 in labels with no predicted samples.

# Visualize the decision tree (not that great because everyone is predicted to be normal)
features = list(X.columns)
labels = ['AllBaby', 'Copy', 'Misc', 'Random', 'SomeBaby']

with open("exif_tree.dot", 'w') as f:
    f = tree.export_graphviz(tree_clf, feature_names=features, class_names=labels, filled=True, rounded=True, out_file=f)
graph = pydotplus.graph_from_dot_data(open('exif_tree.dot').read()) 
Image(graph.create_png())

subject_tree_predict = tree_clf.predict(X)
exif_df['TreePredict'] = subject_tree_predict
exif_df = exif_df[['TreePredict', 'Subject', 'Subject_L', 'ImagePath', 'DateTime', 'Hour', 'TimeBin', 'TimeBin_L', 'Model', 
                   'Model_L', 'FocalLengthIn35mmFilm', 'ExposureProgram', 'ExposureBiasValue', 'Flash', 'Dof', 'Av', 'Tv', 
                   'Sv', 'Bv', 'BrightnessBin', 'BrightnessBin_L']]

pd.options.display.max_colwidth = 500  # display full image path
exif_df[exif_df['Subject']=='SomeBaby'].tail(500)

exif_df.to_csv('exif_predicted.csv')

Yay, we found a cute baby amongst the 'SomeBaby' label!!!¶

import PIL.Image
from PIL import Image

# using a smaller version to prevent huge notebook size
path_to_file = 'C:\\Users\\shera\\Pictures\\2018\\180426-Tot-Lot-ZAK\\180426-185843s.jpg'
baby_img = Image.open(path_to_file)  # use the image library to open file
baby_img

Other models Tested¶

Perceptron¶

# Perceptron does a pretty good job but decision tree is the best

from sklearn.linear_model import Perceptron

perceptron = Perceptron(fit_intercept=False)
perceptron.fit(X_train,y_train)

subject_perceptron_predict = perceptron.predict(X_test)

print(accuracy_score(y_test, subject_perceptron_predict))
print()
print(confusion_matrix(y_test, subject_perceptron_predict))
print()
print(classification_report(y_test, subject_perceptron_predict))

0.6163601775523145

[[1260    0   35    1  597]
 [   3    0    6    0   16]
 [  12    0  121    0  127]
 [   6    0   18    1   97]
 [ 247    0   44    1  562]]

             precision    recall  f1-score   support

    AllBaby       0.82      0.67      0.74      1893
       Copy       0.00      0.00      0.00        25
       Misc       0.54      0.47      0.50       260
     Random       0.33      0.01      0.02       122
   SomeBaby       0.40      0.66      0.50       854

avg / total       0.66      0.62      0.62      3154

C:\Users\shera\Anaconda3\lib\site-packages\sklearn\linear_model\stochastic_gradient.py:128: FutureWarning:

max_iter and tol parameters have been added in <class 'sklearn.linear_model.perceptron.Perceptron'> in 0.19. If both are left unset, they default to max_iter=5 and tol=None. If tol is not None, max_iter defaults to max_iter=1000. From 0.21, default max_iter will be 1000, and default tol will be 1e-3.

C:\Users\shera\Anaconda3\lib\site-packages\sklearn\metrics\classification.py:1135: UndefinedMetricWarning:

Precision and F-score are ill-defined and being set to 0.0 in labels with no predicted samples.

Naive Bayes Classification¶

# Gives a pretty poor result

from sklearn.naive_bayes import GaussianNB

gaussian = GaussianNB()
gaussian.fit(X_train,y_train)

subject_gausian_predict = gaussian.predict(X_test)

print(accuracy_score(y_test, subject_gausian_predict))
print()
print(confusion_matrix(y_test, subject_gausian_predict))
print()
print(classification_report(y_test, subject_gausian_predict))

0.32117945466074826

[[364 519   1  13 996]
 [  0  24   1   0   0]
 [ 12  67  88  16  77]
 [  0  59   7  43  13]
 [ 49 221  26  64 494]]

             precision    recall  f1-score   support

    AllBaby       0.86      0.19      0.31      1893
       Copy       0.03      0.96      0.05        25
       Misc       0.72      0.34      0.46       260
     Random       0.32      0.35      0.33       122
   SomeBaby       0.31      0.58      0.41       854

avg / total       0.67      0.32      0.35      3154

Kmeans¶

# Kmeans confirms that there are 3 clusters (copy and random are not adding anything to the models)

from sklearn.cluster import MiniBatchKMeans

error_costs = []
clusters = range(1,15)  # range of number of clusters to try

for i in clusters:
    kmeansmb_model = MiniBatchKMeans(n_clusters=i)  # i clusters
    kmeansmb = kmeansmb_model.fit(X)
    error_costs.append(kmeansmb.inertia_)

plt.scatter(clusters, error_costs, c = "blue")
plt.plot(clusters, error_costs, c = "red")
plt.title("Error Cost for different numbers of clusters")
plt.xlabel("Number of Clusters")
plt.ylabel("Error Cost (Inertia)")
plt.show()

	ApertureValue	Artist	BitsPerSample	BodySerialNumber	BrightnessValue	CameraOwnerName	ColorSpace	ComponentsConfiguration	CompressedBitsPerPixel	Contrast	Copyright	CustomRendered	DateTime	DateTimeDigitized	DateTimeOriginal	DigitalZoomRatio	ExifImageHeight	ExifImageWidth	ExifInteroperabilityOffset	ExifOffset	ExifVersion	ExposureBiasValue	ExposureMode	ExposureProgram	ExposureTime	FNumber	FileSource	Flash	FlashPixVersion	FocalLength	FocalLengthIn35mmFilm	FocalPlaneResolutionUnit	FocalPlaneXResolution	FocalPlaneYResolution	GPSInfo	ISOSpeedRatings	ImageDescription	ImageLength	ImagePath	ImageUniqueID	ImageWidth	LensMake	LensModel	LensSerialNumber	LensSpecification	LightSource	Make	MaxApertureValue	MeteringMode	Model	Orientation	PhotometricInterpretation	PrintImageMatching	ProcessingSoftware	Rating	RatingPercent	ResolutionUnit	SamplesPerPixel	Saturation	SceneCaptureType	SceneType	SensingMethod	Sharpness	ShutterSpeedValue	Software	SubjectDistance	SubjectDistanceRange	SubjectLocation	SubsecTime	SubsecTimeDigitized	SubsecTimeOriginal	WhiteBalance	XResolution	YCbCrPositioning	YResolution
0	NaN	NaN	NaN	NaN	NaN	NaN	1.0	NaN	NaN	NaN	NaN	NaN	NaN	NaN	2017:08:19 21:20:08	NaN	1535.0	2046.0	250.0	86	b'0220'	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	(6737101, 262144)	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	C:\Users\shera\Pictures\2018\171230-ZAK-7TH-MO...	a97080cd020608d30000000000000000	NaN	NaN	NaN	NaN	NaN	NaN	SONY	NaN	NaN	DSC-RX100M4	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	Picasa	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN
1	NaN	NaN	NaN	NaN	(1984, 2560)	NaN	1.0	b'\x01\x02\x03\x00'	(5, 1)	0.0	NaN	0.0	2017:12:30 14:46:47	2017:12:30 14:46:47	2017:12:30 14:46:47	(16, 16)	3648.0	5472.0	34230.0	290	b'0230'	(3, 10)	0.0	2.0	(1, 125)	(28, 10)	b'\x03'	16.0	b'0100'	(1410, 100)	38.0	NaN	NaN	NaN	NaN	4000.0		NaN	C:\Users\shera\Pictures\2018\171230-ZAK-7TH-MO...	NaN	NaN	NaN	NaN	NaN	((880, 100), (2570, 100), (18, 10), (28, 10))	0.0	SONY	(760, 256)	5.0	DSC-RX100M4	1.0	NaN	b'PrintIM\x000300\x00\x00\x02\x00\x02\x00\x01\...	NaN	NaN	NaN	2.0	NaN	0.0	0.0	b'\x01'	NaN	0.0	NaN	DSC-RX100M4 v1.30	NaN	NaN	NaN	NaN	NaN	NaN	0.0	(350, 1)	2.0	(350, 1)
2	NaN	NaN	NaN	NaN	(1984, 2560)	NaN	1.0	b'\x01\x02\x03\x00'	(5, 1)	0.0	NaN	0.0	2017:12:30 14:46:47	2017:12:30 14:46:47	2017:12:30 14:46:47	(16, 16)	3648.0	5472.0	34230.0	290	b'0230'	(3, 10)	0.0	2.0	(1, 125)	(28, 10)	b'\x03'	16.0	b'0100'	(1410, 100)	38.0	NaN	NaN	NaN	NaN	4000.0		NaN	C:\Users\shera\Pictures\2018\171230-ZAK-7TH-MO...	NaN	NaN	NaN	NaN	NaN	((880, 100), (2570, 100), (18, 10), (28, 10))	0.0	SONY	(760, 256)	5.0	DSC-RX100M4	1.0	NaN	b'PrintIM\x000300\x00\x00\x02\x00\x02\x00\x01\...	NaN	NaN	NaN	2.0	NaN	0.0	0.0	b'\x01'	NaN	0.0	NaN	DSC-RX100M4 v1.30	NaN	NaN	NaN	NaN	NaN	NaN	0.0	(350, 1)	2.0	(350, 1)
3	NaN	NaN	NaN	NaN	(1984, 2560)	NaN	1.0	b'\x01\x02\x03\x00'	(5, 1)	0.0	NaN	0.0	2017:12:30 14:46:49	2017:12:30 14:46:49	2017:12:30 14:46:49	(16, 16)	3648.0	5472.0	34230.0	290	b'0230'	(3, 10)	0.0	2.0	(1, 125)	(28, 10)	b'\x03'	16.0	b'0100'	(1410, 100)	38.0	NaN	NaN	NaN	NaN	4000.0		NaN	C:\Users\shera\Pictures\2018\171230-ZAK-7TH-MO...	NaN	NaN	NaN	NaN	NaN	((880, 100), (2570, 100), (18, 10), (28, 10))	0.0	SONY	(760, 256)	5.0	DSC-RX100M4	1.0	NaN	b'PrintIM\x000300\x00\x00\x02\x00\x02\x00\x01\...	NaN	NaN	NaN	2.0	NaN	0.0	0.0	b'\x01'	NaN	0.0	NaN	DSC-RX100M4 v1.30	NaN	NaN	NaN	NaN	NaN	NaN	0.0	(350, 1)	2.0	(350, 1)
4	NaN	NaN	NaN	NaN	(1834, 2560)	NaN	1.0	b'\x01\x02\x03\x00'	(5, 1)	0.0	NaN	0.0	2017:12:30 14:46:53	2017:12:30 14:46:53	2017:12:30 14:46:53	(16, 16)	3648.0	5472.0	34230.0	290	b'0230'	(3, 10)	0.0	2.0	(1, 125)	(28, 10)	b'\x03'	16.0	b'0100'	(1510, 100)	41.0	NaN	NaN	NaN	NaN	4000.0		NaN	C:\Users\shera\Pictures\2018\171230-ZAK-7TH-MO...	NaN	NaN	NaN	NaN	NaN	((880, 100), (2570, 100), (18, 10), (28, 10))	0.0	SONY	(760, 256)	5.0	DSC-RX100M4	1.0	NaN	b'PrintIM\x000300\x00\x00\x02\x00\x02\x00\x01\...	NaN	NaN	NaN	2.0	NaN	0.0	0.0	b'\x01'	NaN	0.0	NaN	DSC-RX100M4 v1.30	NaN	NaN	NaN	NaN	NaN	NaN	0.0	(350, 1)	2.0	(350, 1)

	Subject	Subject_L	ImagePath	DateTime	Hour	TimeBin	Model	Model_L	FocalLengthIn35mmFilm	ExposureProgram	ExposureBiasValue	Flash	Dof	Av	Tv	Sv	Bv	BrightnessBin	BrightnessBin_L
12611	Random	3	C:\Users\shera\Pictures\2018\999999-Random\180...	2018-04-28 10:37:56	10.0	Daytime	Pixel XL	2	26.0	2.0	0.00000	0.0	1.176592	2.0	10.727	4.000	8.727	Weak Sun	5
12612	Random	3	C:\Users\shera\Pictures\2018\999999-Random\180...	2018-04-28 12:03:27	12.0	Daytime	Pixel XL	2	26.0	2.0	0.00000	0.0	1.176592	2.0	9.400	4.000	7.400	Weak Sun	5
12613	Random	3	C:\Users\shera\Pictures\2018\999999-Random\180...	2018-04-29 14:02:20	14.0	Daytime	Canon EOS 6D	0	84.0	2.0	0.66667	1.0	0.055426	4.0	5.907	7.000	2.907	Flash	2
12614	Random	3	C:\Users\shera\Pictures\2018\999999-Random\180...	2018-04-29 15:06:54	15.0	Daytime	Pixel XL	2	26.0	2.0	0.00000	0.0	1.176592	2.0	12.086	4.986	9.100	Strong Sun	4
12615	Random	3	C:\Users\shera\Pictures\2018\999999-Random\180...	2018-04-29 15:07:05	15.0	Daytime	Pixel XL	2	26.0	2.0	0.00000	0.0	1.176592	2.0	12.025	4.163	9.862	Strong Sun	4

	Subject_L	Hour	TimeBin_L	Model_L	FocalLengthIn35mmFilm	ExposureProgram	ExposureBiasValue	Flash	Dof	Av	Tv	Sv	Bv	BrightnessBin_L
Subject_L	1.000000	0.114506	-0.090914	0.160039	-0.271444	-0.014616	-0.155953	-0.135198	0.065946	-0.156226	0.060998	-0.152811	0.104124	0.212607
Hour	0.114506	1.000000	-0.109333	-0.272586	0.127160	-0.009561	0.232577	0.173850	-0.101463	0.122478	-0.250555	0.366260	-0.358081	0.168964
TimeBin_L	-0.090914	-0.109333	1.000000	0.025781	-0.087849	-0.015981	-0.029968	-0.046293	-0.042907	-0.139662	-0.197367	0.173138	-0.252999	-0.020088
Model_L	0.160039	-0.272586	0.025781	1.000000	-0.790205	-0.052002	-0.806654	-0.489260	0.445900	-0.789533	-0.045884	-0.551639	0.197924	0.039091
FocalLengthIn35mmFilm	-0.271444	0.127160	-0.087849	-0.790205	1.000000	0.081929	0.594621	0.562925	-0.295731	0.788600	0.052754	0.329691	-0.031424	-0.081852
ExposureProgram	-0.014616	-0.009561	-0.015981	-0.052002	0.081929	1.000000	0.016190	0.053219	-0.010409	0.074993	0.015175	-0.011513	0.033344	0.001632
ExposureBiasValue	-0.155953	0.232577	-0.029968	-0.806654	0.594621	0.016190	1.000000	0.341942	-0.382956	0.611385	0.029009	0.557162	-0.252081	-0.056906
Flash	-0.135198	0.173850	-0.046293	-0.489260	0.562925	0.053219	0.341942	1.000000	-0.150610	0.433673	-0.052237	0.009980	0.070905	-0.082654
Dof	0.065946	-0.101463	-0.042907	0.445900	-0.295731	-0.010409	-0.382956	-0.150610	1.000000	-0.233688	-0.084069	-0.345801	0.160054	-0.019696
Av	-0.156226	0.122478	-0.139662	-0.789533	0.788600	0.074993	0.611385	0.433673	-0.233688	1.000000	0.128914	0.300051	0.076122	0.053087
Tv	0.060998	-0.250555	-0.197367	-0.045884	0.052754	0.015175	0.029009	-0.052237	-0.084069	0.128914	1.000000	-0.374526	0.774477	0.154170
Sv	-0.152811	0.366260	0.173138	-0.551639	0.329691	-0.011513	0.557162	0.009980	-0.345801	0.300051	-0.374526	1.000000	-0.840914	0.008416
Bv	0.104124	-0.358081	-0.252999	0.197924	-0.031424	0.033344	-0.252081	0.070905	0.160054	0.076122	0.774477	-0.840914	1.000000	0.078531
BrightnessBin_L	0.212607	0.168964	-0.020088	0.039091	-0.081852	0.001632	-0.056906	-0.082654	-0.019696	0.053087	0.154170	0.008416	0.078531	1.000000

	Subject_L	Hour	TimeBin_L	Model_L	FocalLengthIn35mmFilm	ExposureProgram	ExposureBiasValue	Flash	Dof	Av	Tv	Sv	Bv	BrightnessBin_L
count	12616.000000	12616.000000	12616.000000	12616.000000	12616.000000	12616.000000	12616.000000	12616.000000	12616.000000	12616.000000	12616.000000	12616.000000	12616.000000	12616.000000
mean	1.371909	14.946814	0.944673	1.305644	50.434052	2.001823	0.208870	0.117866	0.760051	2.628456	6.346547	8.216722	0.758281	2.275761
std	1.771063	3.866083	1.257097	0.578066	32.513889	0.066004	0.199988	0.322462	1.347618	0.670207	1.324523	2.084761	2.829936	1.219593
min	0.000000	0.000000	0.000000	0.000000	24.000000	1.000000	0.000000	0.000000	0.055406	1.696000	1.000000	4.000000	-7.626000	0.000000
25%	0.000000	12.000000	0.000000	1.000000	26.000000	2.000000	0.000000	0.000000	0.305049	2.000000	5.322000	6.824000	-1.255500	1.000000
50%	0.000000	16.000000	0.000000	1.000000	39.000000	2.000000	0.300000	0.000000	0.305282	2.971000	6.322000	8.322000	0.610000	2.000000
75%	4.000000	18.000000	1.000000	2.000000	70.000000	2.000000	0.300000	0.000000	1.176592	2.971000	6.966000	10.000000	2.000000	3.000000
max	4.000000	23.000000	4.000000	2.000000	200.000000	4.000000	1.300000	1.000000	10.890400	5.656000	16.610000	13.000000	11.910000	5.000000

	Hour	Model_L	FocalLengthIn35mmFilm	ExposureProgram	ExposureBiasValue	Dof	Av	Tv	Sv	Bv	BrightnessBin_L
0	14.0	1	38.0	2.0	0.3	0.303455	2.971	6.966	10.322	-0.385	3
1	14.0	1	38.0	2.0	0.3	0.303455	2.971	6.966	10.322	-0.385	3
2	14.0	1	38.0	2.0	0.3	0.303455	2.971	6.966	10.322	-0.385	3

	TreePredict	Subject	Subject_L	ImagePath	DateTime	Hour	TimeBin	TimeBin_L	Model	Model_L	FocalLengthIn35mmFilm	ExposureProgram	Dof	Av	Tv	Sv	Bv	BrightnessBin	BrightnessBin_L
8978	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_084514.jpg	2018-04-02 08:45:14	8.0	Morning	3	Pixel XL	2	26.0	2.0	1.176592	2.000	6.906	5.566	3.340	Dusk	1
8979	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_084518.jpg	2018-04-02 08:45:18	8.0	Morning	3	Pixel XL	2	26.0	2.0	1.176592	2.000	6.906	4.379	4.527	Dusk	1
8980	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_084557.jpg	2018-04-02 08:45:57	8.0	Morning	3	Pixel XL	2	26.0	2.0	1.176592	2.000	6.906	4.214	4.692	Dusk	1
8981	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_084600.jpg	2018-04-02 08:46:00	8.0	Morning	3	Pixel XL	2	26.0	2.0	1.176592	2.000	6.906	4.214	4.692	Dusk	1
8982	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_084615.jpg	2018-04-02 08:46:15	8.0	Morning	3	Pixel XL	2	26.0	2.0	1.176592	2.000	6.906	4.214	4.692	Dusk	1
8983	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_094156.jpg	2018-04-02 09:41:56	9.0	Morning	3	Pixel XL	2	26.0	2.0	1.176592	2.000	5.322	6.832	0.490	Dusk	1
8984	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_094157.jpg	2018-04-02 09:41:57	9.0	Morning	3	Pixel XL	2	26.0	2.0	1.176592	2.000	5.907	7.081	0.826	Dusk	1
8985	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_095304.jpg	2018-04-02 09:53:04	9.0	Morning	3	Pixel XL	2	26.0	2.0	1.176592	2.000	5.907	6.395	1.512	Dusk	1
8986	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_095306.jpg	2018-04-02 09:53:06	9.0	Morning	3	Pixel XL	2	26.0	2.0	1.176592	2.000	5.907	6.864	1.043	Dusk	1
8987	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_095310.jpg	2018-04-02 09:53:10	9.0	Morning	3	Pixel XL	2	26.0	2.0	1.176592	2.000	5.907	6.269	1.638	Dusk	1
8988	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_095315.jpg	2018-04-02 09:53:15	9.0	Morning	3	Pixel XL	2	26.0	2.0	1.176592	2.000	5.907	6.014	1.893	Dusk	1
8989	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_095317.jpg	2018-04-02 09:53:17	9.0	Morning	3	Pixel XL	2	26.0	2.0	1.176592	2.000	5.907	6.521	1.386	Dusk	1
8990	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_095320.jpg	2018-04-02 09:53:20	9.0	Morning	3	Pixel XL	2	26.0	2.0	1.176592	2.000	5.907	6.651	1.256	Dusk	1
8991	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_095328.jpg	2018-04-02 09:53:28	9.0	Morning	3	Pixel XL	2	26.0	2.0	1.176592	2.000	5.907	6.395	1.512	Dusk	1
8992	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_103155.jpg	2018-04-02 10:31:55	10.0	Daytime	0	Pixel XL	2	26.0	2.0	1.176592	2.000	6.906	6.590	2.316	Dusk	1
8993	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_103158.jpg	2018-04-02 10:31:58	10.0	Daytime	0	Pixel XL	2	26.0	2.0	1.176592	2.000	6.906	6.459	2.447	Dusk	1
8994	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_103201.jpg	2018-04-02 10:32:01	10.0	Daytime	0	Pixel XL	2	26.0	2.0	1.176592	2.000	6.906	6.459	2.447	Dusk	1
8995	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_104658.jpg	2018-04-02 10:46:58	10.0	Daytime	0	Pixel XL	2	26.0	2.0	1.176592	2.000	6.906	5.993	2.913	Dusk	1
8996	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_104700.jpg	2018-04-02 10:47:00	10.0	Daytime	0	Pixel XL	2	26.0	2.0	1.176592	2.000	6.906	5.176	3.730	Dusk	1
8997	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_104703.jpg	2018-04-02 10:47:03	10.0	Daytime	0	Pixel XL	2	26.0	2.0	1.176592	2.000	6.906	5.604	3.302	Dusk	1
8998	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_104705.jpg	2018-04-02 10:47:05	10.0	Daytime	0	Pixel XL	2	26.0	2.0	1.176592	2.000	6.906	5.903	3.003	Dusk	1
8999	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_104707.jpg	2018-04-02 10:47:07	10.0	Daytime	0	Pixel XL	2	26.0	2.0	1.176592	2.000	6.906	4.926	3.980	Dusk	1
9000	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_104710.jpg	2018-04-02 10:47:10	10.0	Daytime	0	Pixel XL	2	26.0	2.0	1.176592	2.000	6.906	5.993	2.913	Dusk	1
9001	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_110530.jpg	2018-04-02 11:05:30	11.0	Daytime	0	Pixel XL	2	26.0	2.0	1.176592	2.000	6.906	5.669	3.237	Dusk	1
9002	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_110542.jpg	2018-04-02 11:05:42	11.0	Daytime	0	Pixel XL	2	26.0	2.0	1.176592	2.000	6.906	5.696	3.210	Dusk	1
9003	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_110545.jpg	2018-04-02 11:05:45	11.0	Daytime	0	Pixel XL	2	26.0	2.0	1.176592	2.000	6.906	5.632	3.274	Dusk	1
9004	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_110632.jpg	2018-04-02 11:06:32	11.0	Daytime	0	Pixel XL	2	26.0	2.0	1.176592	2.000	6.906	5.263	3.643	Dusk	1
9005	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_110634.jpg	2018-04-02 11:06:34	11.0	Daytime	0	Pixel XL	2	26.0	2.0	1.176592	2.000	6.906	5.176	3.730	Dusk	1
9006	SomeBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_110706.jpg	2018-04-02 11:07:06	11.0	Daytime	0	Pixel XL	2	26.0	2.0	1.176592	2.000	8.460	4.138	6.322	Weak Sun	5
9007	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_110857.jpg	2018-04-02 11:08:57	11.0	Daytime	0	Pixel XL	2	26.0	2.0	1.176592	2.000	5.907	6.736	1.171	Dusk	1
9008	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_110859.jpg	2018-04-02 11:08:59	11.0	Daytime	0	Pixel XL	2	26.0	2.0	1.176592	2.000	5.907	6.996	0.911	Dusk	1
9009	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_110901.jpg	2018-04-02 11:09:01	11.0	Daytime	0	Pixel XL	2	26.0	2.0	1.176592	2.000	5.907	6.526	1.381	Dusk	1
9010	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_115918.jpg	2018-04-02 11:59:18	11.0	Daytime	0	Pixel XL	2	26.0	2.0	1.176592	2.000	5.322	7.050	0.272	Dusk	1
9011	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_115921.jpg	2018-04-02 11:59:21	11.0	Daytime	0	Pixel XL	2	26.0	2.0	1.176592	2.000	5.322	6.960	0.362	Dusk	1
9012	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_115926.jpg	2018-04-02 11:59:26	11.0	Daytime	0	Pixel XL	2	26.0	2.0	1.176592	2.000	5.322	7.084	0.238	Dusk	1
9013	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_122721.jpg	2018-04-02 12:27:21	12.0	Daytime	0	Pixel XL	2	26.0	2.0	1.176592	2.000	6.906	5.926	2.980	Dusk	1
9014	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_123417.jpg	2018-04-02 12:34:17	12.0	Daytime	0	Pixel XL	2	26.0	2.0	1.176592	2.000	6.906	4.310	4.596	Dusk	1
9015	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_123423.jpg	2018-04-02 12:34:23	12.0	Daytime	0	Pixel XL	2	26.0	2.0	1.176592	2.000	7.366	4.401	4.965	Dusk	1
9016	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_123429.jpg	2018-04-02 12:34:29	12.0	Daytime	0	Pixel XL	2	26.0	2.0	1.176592	2.000	6.906	5.251	3.655	Dusk	1
9017	AllBaby	SomeBaby	4	C:\Users\shera\Pictures\2018\180330-DC-Easter-Zak\180402-DC-Easter_123434.jpg	2018-04-02 12:34:34	12.0	Daytime	0	Pixel XL	2	26.0	2.0	1.176592	2.000	7.571	4.956	4.615	Dusk	1